Multicontextual Machine-Learning Approach to Modeling Traffic Impact of Urban Highway Work Zones

Abstract

Impact assessments of highway construction work zones (CWZs) are mandated for all federally funded highway infrastructure improvement projects. However, most existing approaches are ad hoc or project specific, so they are incapable of being benchmarked for any particular spatial region. A novel multicontextual approach to modeling the traffic impact of urban highway CWZs is proposed and tested in this paper. The proposed approach is unique because it models the impact of CWZ operations through a multicontextual quantitative method using big data for improved accuracy. In this study, a machine-learning technique was adopted to predict long-term traffic flow rates and the corresponding truck percentages. With the use of these predicted values, stereotypical patterns of traffic volume-to-capacity ratios were created for typical urban nighttime closures. Third-order curve-fitting models to achieve potential work zone travel time delays in heavily trafficked large urban cores were then developed and validated. This study will greatly help state and local governments and the general traveling public in major cities know the potential traffic flow resulting from construction and thereby facilitate progress on highway improvement projects with the better-informed work zone traffic flow and thus improve safety and mobility in and between CWZs.